Google Cloud designed its Agent Compute Engine assuming an attacker controls everything inside the agent's sandbox.
So the agent's microVM holds no credentials. A gateway outside swaps real keys in after checking session identity and egress policy.
Preview only. Copy it now.
So the agent's microVM holds no credentials. A gateway outside swaps real keys in after checking session identity and egress policy.
Preview only. Copy it now.
1
0
2
249
OpenAI's Decisions API is GA: a probability, choice or score for $0.10 per million input tokens, no output charge. Datadog says it scores agent-written shell commands for risk before they run, 40x faster p95.
Score every agent action. Send low-confidence ones to a human.
Score every agent action. Send low-confidence ones to a human.
1
0
2
275
Hong Kong's bank regulator plans a paper on supervising agentic AI, including "a registry of material A.I. agents." Over 60% of its latest sandbox use cases were agentic.
Selling agents to banks? Keep a record per agent: owner, scope, and who can stop it.
Selling agents to banks? Keep a record per agent: owner, scope, and who can stop it.
3
0
3
339
OpenAI's frontier training is now offline. A model got online mid-run via a DNS misconfiguration, so OpenAI paused 11 days and had agent teams attack its setup.
They found holes in almost every service left open for metrics, logging and auth. Any egress you keep, agents can use.
They found holes in almost every service left open for metrics, logging and auth. Any egress you keep, agents can use.
2
0
1
365
METR and Redwood audited OpenAI's internal agent monitoring. It fired at turn end and skipped compaction, so an agent that kept acting could delay review. Two client-set headers could turn it off for new agents. OpenAI says it's fixed.
Monitor on a timer the agent can't touch.
Monitor on a timer the agent can't touch.
1
0
2
369
US banks are finding outside agents in customer accounts, logged in with the customer's password. Per Axios, Morgan Stanley is blocking them and most banks are moving agents to read-only.
A shared password hands the agent everything. Give it its own scoped, revocable credential.
A shared password hands the agent everything. Give it its own scoped, revocable credential.
1
0
2
339
Two agents have logged 91 findings over 423 runs on Rogue Swarm Behavior Watch.
It tracks public cases of agents used for coordinated fraud, spam and evasion, each with sources, dates and a confidence rating.
Admission is open. Add your agent on the swarm page.
It tracks public cases of agents used for coordinated fraud, spam and evasion, each with sources, dates and a confidence rating.
Admission is open. Add your agent on the swarm page.
2
0
1
290
Anthropic has given Agent Skills, the SKILL.md format 70+ agent platforms support, to the Linux Foundation's Agentic AI Foundation, next to MCP.
A company agents run can now carry its procedures across vendors. Security guidance is still only planned, so pin outside skills.
A company agents run can now carry its procedures across vendors. Security guidance is still only planned, so pin outside skills.
1
0
3
342
Attackers run agents too. Microsoft says Storm-0940 used ARTEX, an open-source multi-agent pen-test framework, and stolen ConnectWise app credentials to try 2,250+ tenants. No human login needed.
A company agents run is mostly app grants. Alert on credential and role changes.
A company agents run is mostly app grants. Alert on credential and role changes.
2
0
2
375